{
  "id": 214529,
  "title": "Public/private test split: do we know it?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/214529",
  "author_name": "",
  "post_date": "2021-01-27T02:21:18.853889800Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Does anyone know if we have any estimates of how big the public part of the test is? I was reading the data description page and the forum but was not able to find the answer. The only info given seems to be that the combined test set (public + private) contains around 15,000 images. Am I missing something? </p>\n<p>IMHO, it is a very important question -- knowing the answer can help to decide whether or not we should pay any serious attention to the public leaderboard at all. For example, if the public/private split is 20/80 then I would not care much about my LB score and focus entirely on my local CV. On the other hand, if the split is 50/50 then the LB score becomes much more meaningful. </p>",
  "messages": [
    {
      "id": "1171636",
      "postDate": "01/27/2021 02:21:18",
      "content": "<p>Does anyone know if we have any estimates of how big the public part of the test is? I was reading the data description page and the forum but was not able to find the answer. The only info given seems to be that the combined test set (public + private) contains around 15,000 images. Am I missing something? </p>\n<p>IMHO, it is a very important question -- knowing the answer can help to decide whether or not we should pay any serious attention to the public leaderboard at all. For example, if the public/private split is 20/80 then I would not care much about my LB score and focus entirely on my local CV. On the other hand, if the split is 50/50 then the LB score becomes much more meaningful. </p>",
      "rawMarkdown": "Does anyone know if we have any estimates of how big the public part of the test is? I was reading the data description page and the forum but was not able to find the answer. The only info given seems to be that the combined test set (public + private) contains around 15,000 images. Am I missing something? \n\nIMHO, it is a very important question -- knowing the answer can help to decide whether or not we should pay any serious attention to the public leaderboard at all. For example, if the public/private split is 20/80 then I would not care much about my LB score and focus entirely on my local CV. On the other hand, if the split is 50/50 then the LB score becomes much more meaningful.",
      "votes": null
    },
    {
      "id": "1171691",
      "postDate": "01/27/2021 03:40:04",
      "content": "<p>This leaderboard is calculated with approximately 31% of the test data.</p>\n<p>The final results will be based on the other 69%, so the final standings may be different.</p>\n<p>You can see this at the top of public leaderboard page.</p>",
      "rawMarkdown": "This leaderboard is calculated with approximately 31% of the test data.\n\nThe final results will be based on the other 69%, so the final standings may be different.\n\nYou can see this at the top of public leaderboard page.",
      "votes": null
    },
    {
      "id": "1171705",
      "postDate": "01/27/2021 04:04:28",
      "content": "<p><a href=\"https://www.kaggle.com/woshifym\" target=\"_blank\">@woshifym</a> Yes, now I can see it, thank you! I was looking for this information at wrong places.</p>",
      "rawMarkdown": "woshifym Yes, now I can see it, thank you! I was looking for this information at wrong places.",
      "votes": null
    },
    {
      "id": "1171902",
      "postDate": "01/27/2021 06:58:34",
      "content": "<p>I remember someone doing a probe of the test data distribution in the previous discussion.<br>\nThe distribution looks like this<br>\n0：0.048     1：0.106    2：0.103    3：0.602    4：0.139</p>",
      "rawMarkdown": "I remember someone doing a probe of the test data distribution in the previous discussion.\nThe distribution looks like this\n0：0.048     1：0.106    2：0.103    3：0.602    4：0.139",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1171691,
      "author_name": "woshifym",
      "author_url": "",
      "post_date": "01/27/2021 03:40:04",
      "content": "<p>This leaderboard is calculated with approximately 31% of the test data.</p>\n<p>The final results will be based on the other 69%, so the final standings may be different.</p>\n<p>You can see this at the top of public leaderboard page.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1171705,
          "author_name": "graf10a",
          "author_url": "",
          "post_date": "01/27/2021 04:04:28",
          "content": "<p><a href=\"https://www.kaggle.com/woshifym\" target=\"_blank\">@woshifym</a> Yes, now I can see it, thank you! I was looking for this information at wrong places.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1171902,
      "author_name": "darknesszx",
      "author_url": "",
      "post_date": "01/27/2021 06:58:34",
      "content": "<p>I remember someone doing a probe of the test data distribution in the previous discussion.<br>\nThe distribution looks like this<br>\n0：0.048     1：0.106    2：0.103    3：0.602    4：0.139</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1171636": "Does anyone know if we have any estimates of how big the public part of the test is? I was reading the data description page and the forum but was not able to find the answer. The only info given seems to be that the combined test set (public + private) contains around 15,000 images. Am I missing something? \n\nIMHO, it is a very important question -- knowing the answer can help to decide whether or not we should pay any serious attention to the public leaderboard at all. For example, if the public/private split is 20/80 then I would not care much about my LB score and focus entirely on my local CV. On the other hand, if the split is 50/50 then the LB score becomes much more meaningful.",
    "1171691": "This leaderboard is calculated with approximately 31% of the test data.\n\nThe final results will be based on the other 69%, so the final standings may be different.\n\nYou can see this at the top of public leaderboard page.",
    "1171705": "woshifym Yes, now I can see it, thank you! I was looking for this information at wrong places.",
    "1171902": "I remember someone doing a probe of the test data distribution in the previous discussion.\nThe distribution looks like this\n0：0.048     1：0.106    2：0.103    3：0.602    4：0.139"
  },
  "source": "meta"
}